Skip to main content

​
Project setup

The platform stores project names, task descriptions, requirements, plans, and team conversations. The conversational guide sends project descriptions to Anthropic. Use it for requirements, not confidential source documents or secrets. The guided form is available without a guide conversation.

​
Data workspace

Company TXT and JSONL imports are available for projects with managed preparation/training. Their text and expected outputs are stored in the hosted data workspace. Imports are blocked for company-data projects configured for private preparation until a private runner is connected. Public starters have a separate import path. Dataset content is not sent to the onboarding guide. PDF/OCR ingestion and third-party data connectors are not available in this release.

​
Inference

Chat and API inputs are forwarded by the hosted platform to the configured serving runtime. Managed demo prompts therefore reach the RunPod worker. A runtime hosted on customer hardware does not imply direct browser-to-customer communication: the current API path passes through the platform. The standalone chat keeps its conversation in page memory. This is not a claim that providers or infrastructure retain no logs. Agree on logging, retention, access, and deletion before using confidential information.

​
Customer-controlled processing

The intended private runner targets Linux servers with NVIDIA GPUs and Docker. Installation and enforcement of a complete private processing boundary are still being built. Choosing Your infrastructure records the requirement; it does not make the current hosted workflow air-gapped or fully private. For a confidential pilot, verify document preparation, training, inference, agent calls, artifact storage, and content-bearing logs as one complete data flow.